![[2505.12641] Single Image Reflection Separation via Dual Prior Interaction Transformer](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2505.12641] Single Image Reflection Separation via Dual Prior Interaction Transformer
Summary
This paper presents a novel approach to single image reflection separation using a Dual Prior Interaction Transformer, enhancing the extraction of transmission and reflection layers from mixed images.
Why It Matters
The research addresses limitations in existing methods for image reflection separation, which often fail to effectively utilize transmission priors. By introducing a dual-prior framework, this work could significantly improve image processing applications in computer vision, impacting fields like photography, augmented reality, and visual effects.
Key Takeaways
- Introduces a Local Linear Correction Network (LLCN) for efficient transmission prior generation.
- Proposes a Dual-Prior Interaction Transformer (DPIT) for deep fusion of general and transmission priors.
- Demonstrates state-of-the-art performance on multiple benchmark datasets.
- Addresses the challenge of modeling transmission priors in complex scenarios.
- Enhances the quality of image reflection separation with minimal parameters.
Computer Science > Computer Vision and Pattern Recognition arXiv:2505.12641 (cs) [Submitted on 19 May 2025 (v1), last revised 14 Feb 2026 (this version, v3)] Title:Single Image Reflection Separation via Dual Prior Interaction Transformer Authors:Yue Huang, Tianle Hu, Yu Chen, Zi'ang Li, Jie Wen, Xiaozhao Fang View a PDF of the paper titled Single Image Reflection Separation via Dual Prior Interaction Transformer, by Yue Huang and 5 other authors View PDF HTML (experimental) Abstract:Single image reflection separation aims to separate the transmission and reflection layers from a mixed image. Existing methods typically combine general priors from pre-trained models with task-specific priors such as text prompts and reflection detection. However, the transmission prior, as the most direct task-specific prior for the target transmission layer, has not been effectively modeled or fully utilized, limiting performance in complex scenarios. To address this issue, we propose a dual-prior interaction framework based on lightweight transmission prior generation and effective prior fusion. First, we design a Local Linear Correction Network (LLCN) that finetunes pre-trained models based on the physical constraint T=SI+B, where S and B represent pixel-wise and channel-wise scaling and bias transformations. LLCN efficiently generates high-quality transmission priors with minimal parameters. Second, we construct a Dual-Prior Interaction Transformer (DPIT) that employs a dual-stream chann...